Copyright 2020 Robert Clark
In the blog post, "About the Lancet paper on hydroxychloroquine and COVID-19", I discussed a letter I sent to the authors of the recent Lancet paper:
Prof Mandeep R Mehra, MD Sapan S Desai, MD Prof Frank Ruschitzka, MD Amit N Patel, MD
Published: May 22, 2020 DOI:https://doi.org/10.1016/S0140-6736(20)31180-6
The issue I had was with the higher number of patients placed on ventilators among the HCQ group compared to the non-HCQ, over twice as many. COVID-19 patients placed on ventilators have poor outcomes. In New York for example, only 20% of them survive it. This would skew the mortality numbers for the HCQ group towards higher mortality.
So the question then was this: was the higher number of intubated patients in the HCQ group seen in this study because of the HCQ or was it because even after the authors applied methods to compensate for HCQ being given more to the sicker patients, this bias against HCQ still remained in the data?
I'm inclined to believe the latter because while HCQ had been connected to heart problems it was not known to cause breathing problems.
If this is the case then how the authors conducted their statistical adjustments to the data becomes extremely important. I advised the authors to release their data and describe their procedures for how they compensated for HCQ being prescribed to the more sicker patients to begin with.
The authors releasing their data and procedures becomes even more important because of this recent news report:
Questions raised over hydroxychloroquine study which caused WHO to halt trials for Covid-19.
The company that provided the data for the Lancet paper has been accused of falsifying data and Australia has denied ever supplying them with the information they claimed comes from Australian hospitals.
This company, Surgisphere, has refused to hand over their data. In view of the importance of this paper with randomized, controlled trials that could have finally determined the validity of HCQ’s effectiveness being cancelled, the authors should be required to either hand over their data or withdraw their paper.
Because of the controversy and doubt about how the data in such studies that are non-randomized are adjusted to mimic randomized studies, I recommend also the simplified approach I discussed in the prior post mentioned above. It's advantages are its transparency and simplicity. Everyone can see how the numbers are being developed at every step of the process, and anyone with just a hand-calculator can do the calculation themselves, not needing advanced statistical packages to do it.
Letter to the authors of the Lancet paper:
Robert G Clark
Wed 5/27/2020 12:12 AM
Thanks for taking the time to respond, Dr. Mehra. Being intubated is a huge factor towards mortality for COVID-19 patients. In New York, only 20% of intubated COVID-19 patients survive:
So it is a key question of whether the HCQ caused the intubation. I’m inclined to think no since HCQ while it was previously known to cause heart problems was not previously known to cause breathing problems. But in view of the questions importance, this is something that should be investigated by interviewing the doctors who treated these intubated patients. Prior to being put on the ventilator was their patients condition so poor that it was to be expected for the intubation to occur whether or not they took HCQ?
Note then since COVID-19 affects the heart also this would explain the higher number of heart problems with the HCQ group if they were more greatly represented among the intubated patients.
In fact, I would also expect for example there would be a higher number of strokes among the HCQ group in your study: the intubated patients being in worse condition would skew those numbers up as well for the HCQ group.
So think of any life threatening medical condition that is known to be caused by COVID-19 but is not known to be caused by HCQ. I would imagine you will find those conditions will still predominate in the HCQ group even after you made your adjustments for HCQ being given to sicker patients.
This uncertainty on the underlying cause of the higher mortality in the HCQ group is why I would like to see my simplified approach to equalizing between the HCQ and non-HCQ groups risk factors be implemented.
Because of the controversy of the issue, many people will have doubts about the conclusions when all they are given are the “adjusted” numbers not the original numbers.
Furthermore, the simplified approach can be followed every step of the way to see the calculations are being done appropriately and can be done by anyone with just a high school understanding of proportions and percentages. You don’t need advanced statistical knowledge or advanced statistical computer packages to do it. The calculations can even be done simply by hand.
To this end I again request that the original numbers be provided so pretty much anyone can make their own calculations for equalizing between the HCQ and non-HCQ groups.
Also, remember the importance of having the original numbers is not just for doing the equalizing calculation. It may very well be that certain risk factors are better addressed by HCQ. You can imagine a scenario, for example, that people with hypertension have better outcome on HCQ. You can only make this comparison to the non-HCQ case if you have the original numbers.
Dept. of Mathematics
One University Place
Chester, PA 19013 USA